AI Data Center Power Fluctuations Damage Equipment

Rapid fluctuations in power demand at AI data centers are causing critical equipment such as batteries, generators, and cooling systems to malfunction or wear out prematurely. The problem stems from the massive and volatile power consumption of AI workloads, particularly during model training, which can cause power usage to spike up to 50% above design capacity. This strain has led to cracked turbines at facilities like Xai's Colossus in Memphis, Tennessee, and premature battery failures at sites worldwide. The reliability issues are adding costs and reducing uptime for data center operators, with some facilities seeing uptime as low as 80%. This comes amid investor concerns about the massive capital expenditures by hyperscalers and the potential for faster-than-expected depreciation of AI infrastructure. The problem also poses risks to grid stability, prompting the North American Electric Reliability Corporation to issue a level-three alert. Companies are developing solutions, including power smoothing equipment and test facilities, to address these challenges.